Attribution systems are useful, but they are not truth machines. They tell the company which touchpoints appeared near a conversion. They do not automatically tell the company whether the spend caused the conversion.
That distinction matters because capital allocation depends on causality. If a customer would have converted anyway, the attributed channel may be collecting credit rather than creating value. This is especially common in retargeting, branded search, affiliate, lifecycle, influencer, and promotion-heavy work.
Attribution answers a credit question. Incrementality answers a counterfactual question. What happened because of the spend that would not have happened without it? The second question is slower and less flattering. It is also closer to the truth growth leaders need.
The danger is not that attribution is fake. The danger is that companies forget what it is allowed to prove. A last-touch or multi-touch model can help organize channel reporting. It cannot, by itself, justify marginal spend. It needs experiments, holdouts, geo tests, suppression tests, or other counterfactual evidence.
An incrementality test plan should identify the channel, the claimed effect, the population, the holdout method, the time window, the success metric, and the decision that will change based on the result. Tests without decision consequences become analytics theater.
Incrementality also has to be net. If a campaign creates customers but those customers have lower contribution margin, higher refunds, weaker retention, or heavier support load, the causal effect is not enough. The relevant question is incremental net payback.
This can create tension because attribution makes teams feel productive. Dashboards fill with credited conversions. Agencies, platforms, and internal teams can all point to numbers that reward continued spend. Incrementality can take those numbers away.
Good growth leaders welcome that discomfort. They would rather know a channel is smaller than believed than keep allocating budget to a comforting fiction. The goal is not to embarrass teams. The goal is to move money toward work that truly changes customer behavior.
The operator test: choose one channel with strong attributed performance and ask what percentage of its conversions would likely happen without the channel. If the answer is a guess, the next budget increase should be paired with a test.
Attribution is a reporting system. Incrementality is an allocation discipline. Confusing the two is one of the fastest ways to turn growth sophistication into spend justification.
The practical challenge is that incrementality usually makes measurement slower. A platform dashboard updates instantly. A holdout test takes time, design, and patience. That timing mismatch creates pressure to manage by whatever number is available soonest, even if the available number is the least causal.
This is why the review should classify channels by evidence quality. Some channels may have strong experimental proof. Some may have directional lift evidence. Some may only have attribution. The budget should reflect that confidence. A channel with weak causal evidence can still receive money, but it should not receive the same trust as one that has survived a counterfactual test.
Incrementality also changes the politics of channel ownership. If a lifecycle team, paid team, affiliate program, or retargeting campaign has been credited with conversions for years, testing the counterfactual can feel threatening. Leaders have to frame the exercise as capital clarity, not a hunt for guilty teams.
The best outcome is not always cutting the channel. Sometimes incrementality work shows that the channel matters less than believed, works only in specific segments, or works only at certain frequencies. That still helps. It turns a blunt budget into a sharper allocation rule.
The uncomfortable example is retargeting. It often looks excellent because it touches people who are already warm. That does not mean it has no value. It may increase conversion speed, reduce abandonment, or recover buyers who would otherwise drift away. But those claims need a counterfactual instead of a platform report with impressive credited revenue.
Branded search has a similar problem. If customers already know the company and search for its name, the paid click may protect against competitors or make navigation easier. It may also be paying for traffic the company would have received anyway. The question is not whether the channel appears in the path. The question is how much behavior changes when the spend is removed or reduced.
This is why incrementality should be treated as a habit, not a one-time audit. Every major credited channel should eventually face some version of a holdout, suppression, geo split, matched-market test, or natural-experiment review. The method can be imperfect. The point is to keep asking whether credited growth is caused growth.
The best teams do not make this precious. They do not wait for a perfect experiment design before questioning a suspicious number. They start with the channels where the gap between credited value and likely caused value feels largest, then build better evidence over time. A rough but honest test is often more useful than another quarter of pretending the attribution dashboard is a court transcript.
There is also a tone shift that matters. Instead of asking "did this channel work?" ask "what did this channel make happen?" That question is harder to dodge. It pushes the team toward behavior change, not credit assignment. It also makes room for nuance: a channel may not create new demand, but it may speed up conversion, defend brand demand, or improve close confidence in a narrow segment. Those are real effects if the company can prove them.
A better monthly review would label channels by proof level. Retargeting might be 'credited, not yet tested.' Branded search might be 'defensive, needs suppression test.' Paid social prospecting might be 'directional lift, weak segment read.' A partner motion might be 'slow but visibly incremental in enterprise accounts.' The labels matter because they stop every channel from pretending to have the same evidence.
The finance implication is straightforward. A dollar supported only by attribution should face a smaller budget increase than a dollar supported by causal evidence. That does not make the attribution dollar useless. It just means the company should buy more proof before it buys a much larger version of the same uncertainty.
The test does not have to be theatrical. Turn off spend in a small region. Hold back a segment for a limited period. Compare similar accounts with and without the touch. Watch what happens to conversion, margin, and cash timing. The result will not be perfect, but it will usually be more honest than another attribution export.
Teams should also write down what would change their minds before the test starts. If a channel needs at least ten percent incremental lift to earn more budget, say that in advance. If it only needs to defend existing demand at a low cost, say that too. Clear stakes make the analysis harder to bend after the results arrive.
This is part 5 of 10 in The Capital Allocation Theory of Growth.